This paper proposes an algorithm for binary-class imbalanced data from a feature selection perspective.Feature selection is a pre-processing steps for data mining.A good feature selection methods can not only cope with data redundancy and over fitting,but also reduce the complexity and the runtime of an algorithm.The main idea of the feature selection algorithm is assigning different weights to majority class and minority class according to the neighborhood rough set theory.The weight of minority class is bigger than that of majority class.The advantage of this algorithm is that it can effectively find the subset which affect the minority class greatly.In this paper,the BPSO algorithm is used to find the feature subset,which can quickly find the optimal solution of the adaptive function without generating all the feature subsets.In addition,the feature selection algorithm is integrated with the classic SMOTE and ENN algorithms to form a hybrid algorithm,which absorbs the results of previous research on imbalanced samples classification,and has better classification performance. |